Method and apparatus for extended depth-of-field image restoration

1. A method of processing a captured image comprising:

receiving a captured image;

dividing the captured image into a plurality of tiles;

reading at least one prestored eigen-kernel from memory;

reading at least one scalar value from memory; and

for each pixel in the tile:

reading image data in an M×M window, where M is a number of pixels;

determining a mean of the windowed data;

estimating a variance of noise at a current pixel using the mean and a noise model;

estimating a variance of the image data in the window;

estimating a variance of an original image taking into account degrading blur, noise, and the variance of the image data; and

determining a signal-to-noise ratio at the current pixel, wherein estimating a variance of an original image is performed according to:

σ x 2 =(σ y 2 −σ n 2 )/φ( h ),

wherein φ(h) is a scalar value representing a strength of the point spread function, σ n 2 is the estimated variance of noise, σ y 2 is the estimated variance of the image data, and σ x 2 is the estimated variance of the original image.

2. The method of claim 1 wherein the scalar value φ(h) is determined according to:

ϕ

⁡

(

h

)

=

γ

⁢

∑

p

⁢

∑

q

⁢

h

2

⁡

(

p

,

q

)

,

wherein γ is a lens-design specific parameter, and h is the PSF for the tile (p,q).

3. A method of processing a captured image comprising:

receiving a captured image;

dividing the captured image into a plurality of tiles;

reading at least one prestored eigen-kernel from memory;

reading at least one scalar value from memory;

for each pixel in the tile:

reading image data in an M×M window, where M is a number of pixels;

determining a mean of the windowed data;

estimating a variance of noise at a current pixel using the mean and a noise model;

estimating a variance of the image data in the window;

estimating a variance of an original image taking into account degrading blur, noise, and the variance of the image data; and

determining a signal-to-noise ratio at the current pixel;

comparing the determined signal-to-noise ratio with a predetermined threshold; and

filtering the captured image signal according to one of:

if the determined signal-to-noise ratio is lower than the threshold, then determining reconstruction coefficients by either preserving a value observed or using a smoothing filter; and

using the determined signal-to-noise ratio at the current pixel as an input to a polynomial representation to determine reconstruction coefficients to be used for determining a restoration kernel appropriate for the current pixel.

4. The method of claim 3 , further comprising:

using the coefficients to linearly combine the eigen-kernels stored for the current tile to determine the restoration kernel that is to be applied at the current pixel;

performing a convolution at the current pixel using the determined restoration kernel; and

replacing a pixel value for the current pixel with the convolved value.

5. The method of claim 3 , further comprising, if the current pixel is located within a tile boundary region, interpolating a value for the pixel.

6. The method of claim 3 , wherein the at least one prestored eigen-kernel is different for each tile.

7. A method of processing a captured image comprising:

receiving a captured image;

dividing the captured image into a plurality of tiles;

dividing each of the plurality of tiles into at least one sub-tile;

reading at least one scalar value from memory;

reading prestored eigen-kernels from memory;

determining a signal-to-noise ratio for the current sub-tile;

for each sub-tile:

using the determined signal-to-noise ratio for the current sub-tile as an input to a polynomial representation to determine reconstruction coefficients to be used for determining a restoration kernel appropriate for the current sub-tile; and

using the coefficients to linearly combine the eigen-kernels stored for the current file to determine a restoration kernel that is to be applied at the current pixel; and

for each pixel in the tile:

reading image data in an M×M window, where M is a number of pixels;

determining a mean of the windowed data;

performing one of:

estimating a variance of noise at a current pixel or reading an estimated variance of noise for the sub-tile from a memory;

determining an estimated variance of the image data in the window; and

determining an estimate of a variance of an original image taking into account both degrading blur and noise acting on the original image.

8. The method of claim 7 , further comprising:

comparing the determined estimated original image variance with the estimated noise variance;

performing a convolution at the current pixel using the determined restoration kernel; and

replacing a pixel values for the current pixel with the convolved value.

9. The method of claim 8 , further comprising if the current pixel is located within a tile boundary region, interpolating a value for the pixel.